A Machine Learning Approach to Classify Network Traffic

There is a significant increase of cloud and networking-enabled applications, leading to an exponential growth of network traffic. Monitoring all these applications and their generated traffic is a challenging and complex task, especially in regard to anonymous networks used to access the Dark Web or darknet. Manual investigation of the network traffic to determine patterns of malicious activity is a very difficult task and the results might not be satisfactory. The usage of machine learning is the most viable approach for the classification of the traffic as normal or malign activity. This paper analyzes CIC-Darknet 2020 dataset to classify the benign and darknet traffic. Before applying any classifiers to our dataset, we have balanced it using Synthetic Minority Oversampling Technique (SMOTE). We have applied PCA to reduce dimensionality, furthermore, ensemble techniques, logistic classifiers, tree-classifiers, and Naive Bayes have been compared and evaluated thoroughly with various evaluation metrics Accuracy, Precision, Recall, F1-Score, and Mathew’s Correlation Coefficient (MCC).